Intra- and inter-frame prediction in bandwidth scalable coding of wideband speech

Abstract

This study discusses a predictive coding scheme that utilises both the intra-frame correlation (here referred to as redundancy between frequency bands) and inter-frame correlation in bandwidth scalable coding of wideband speech. Prediction error filters that combine intra-frame prediction with inter-frame prediction in a cascade connection are examined. Three representative intra-frame functions that have been developed in the area of bandwidth extension (BWE) are considered to see how well they exploit the intra-frame correlation. In addition, some resulting changes are investigated when they are combined with the predictive vector quantiser that exploits the inter-frame correlation. The performances of the Gaussian mixture model-based and hidden Markov model-based BWE functions are similar in the context of intra-frame coding, but both of them significantly outperform the linear BWE function in the same context. However, it is also observed that the linear function is the most promising in the context of the joint predictive coding of intra-frame and inter-frame. These results are supported by some statistical tests.